Paying for intelligence
that still feels dumb?
You're paying for Claude, Copilot, ChatGPT. They're still re-asking what Sarah posted in Slack last Tuesday. Halyard plugs every agent you run into the knowledge your team already has.
Agents don't know what your team knows.
So they guess. Worse, they sound confident and wrong. Ungrounded AI is a hallucination factory that burns the trust you're trying to build.
$ claude "add rate limiting to /api/auth"
Using express-rate-limit with a 100 req / 15 min
window on /auth/login and /auth/signup — industry
standard for auth endpoints.
⚠ team migrated off express 18 months ago
⚠ no citation · no source · no grounding$ claude "add rate limiting to /api/auth"
Your stack uses Fastify + @fastify/rate-limit
per ADR-017 (Sarah, 2024-09).
Auth endpoints share the strict-auth preset
decided in #eng-decisions.
✓ ADR-017 · Notion
✓ #eng-decisions · Slack
✓ fastify.config.ts · GitHub
One knowledge layer.
Every tool your team uses.
Halyard indexes the places your team actually produces knowledge — Slack, Notion, GitHub, Drive, Linear, meetings — and makes it retrievable by whichever AI agent or teammate needs it. Whether you run Claude, Codex, Cursor, or whatever ships next, Halyard's the layer underneath.
Multi-source, not single-silo
Real knowledge lives where work happens — in threads, PRs, and meetings, not just your wiki. Halyard captures it from all of them.
Both humans and agents
When the answer's written down, your agent cites it. When it isn't, Halyard routes the question to the person who knows — and captures the reply for next time.
Living context
Knowledge updates as work happens — not when someone remembers to edit a wiki page. Agents always get the current answer, not last quarter's.
When it's not written down yet,
we make it durable.
An agent hits an unknown. Halyard routes it to the right human in Slack. Their reply doesn't disappear into a thread — it becomes searchable knowledge for every future agent, and every future hire.
We use Fastify + @fastify/rate-limit now. Dropped express-rate-limit in the 2024 migration. See ADR-017 for the why.